Multicolor reordering for computing moments in particle-in-cell plasma simulations

Thread parallelism in the computation of the current density/charge density in particle-in-cell plasma simulations has been performed by using the reduction operation conventionally, which is known to have a larger computational overhead with a larger number of threads. In the present study, two types of multicolor reordering, i.e., loop striding and loop tiling/blocking are introduced for a particle shape function with an arbitrary degree, which is free from the reduction operation. The present performance measurement result suggests that the loop tiling is superior to the loop striding. (C) 2022 Elsevier B.V. All rights reserved.